SQL Release Notes

This page documents the version history of SQL in KDB-X, including released versions, fixes, and improvements.

2026.10.20

• Window functions: lag, lead, nth_value added; row_number is now 1-based (was 0-based)
• Window frames: rows/range/groups between .. and ..; the default frame with order by is now running rather than whole-partition (affects sum/avg/.. and last_value)
• Window functions can be nested in expressions, used with group by, select *, aliased joins and select lists made only of windows
• Unaliased window columns are named after the argument (sum(px) -> px), no longer after the order by column
• Window functions on partitioned tables threw 'rank when order by or partition by used the partition column; like q's sums/mavg, windows restart at each partition rather than carrying over
• values (..) can be queried directly or used as a from source; column alias lists (as t (a, b)) are supported
• Group by can refer to a select-list alias
• show hung the process on kdb-x 2026.10.02 and later

2026.07.15

• pgwire now supports the extended query protocol

2025.12.16

• Improved like, ilike and POSIX match performance

2025.06.27

• Support for .Q.bv

2025.05.22

• Feature flag update

2025.01.16

• select on keyed tables was broken since 2024.02.09

2024.07.30

• Improved Tableau compatibility

2024.05.31

• .s.sp now supports keyed table arguments
• Partitioned table queries accessing linked columns with predicate matching no rows failed with 'link

2024.03.01

• Implemented limit .. offset

2024.02.09

• Improved performance for select on wide tables

2023.11.20

• Query compilation/execution no longer requires full metadata of the tables involved. Only the columns mentioned will be opened, similar to q-sql
• timespan can be cast to numeric types

2023.05.30

• Proper handling of empty queries to prevent JDBC crash
• Comparison of char against string could throw 'type
• ilike now implements case-insensitive like

2023.03.20

• Array indexing works for array-returning functions like current_schemas()

2023.02.24

• Parser support for COPY subprotocol commands, .s.ver bumped to 1.1
• Conversions to type text, including pgwire, now return full data for array types, without literal type suffix
• Metadata retrieval could fail for tables containing dictionary rows
• Evaluation of empty (or comments only) statement threw 'parse
• Default schema public is no longer case sensitive

2022.12.19

• pgwire now supports kdb+ IPC messages of up to 1TB for data returned. Previously threw 'limit for >2GB

• Serialization into pgwire format (done in s.k) is now faster, parallel and memory-efficient, for example:

  q)n:1000000;t:([]a:n?`8;b:til n;c:"f"$til n;d:n#.z.d;e:n#.z.z;f:n#.z.p;g:n#.z.t;h:n#.Q.a)
  q)\ts .s.spf[;1b]t / internal fn to serialize 1b is binary
  1087 134218480     / was: 2528 1069486336
  

• In pgwire, kdb+ types minute and second now mapped to Postgres time, linked columns to bigint, and arrays/nested data to text representation

• Improved compatibility with abInitio, DBeaver
• Fixed display of type char in PowerBI

• Lexer now reports errors, for example:

  q)s)1.2g,1
  'lexer unexpected:g,1
  

2022.11.11

• extract(epoch from time|timespan) now uses 2010.01.01 as date to work around Grafana display bug
• Grafana query builder is able to list tables

• Support for implicit join notation (was previously executed as cross join)

  s)select * from t0,t1,t2 where t0.a=t1.b and t1.c=t2.d
  

2022.10.07

• Improved compatibility with psqlodbc for pgwire. NB: psqlodbc option UseServerSidePrepare=0 is necessary for queries with parameters

2022.10.03

• select without aggregates was slow
• Browsing tables in Spotfire information designer produced errors
• pgwire text format now has 1us precision timestamp
• q type minute now mapped to pgwire time

2022.08.03

• Compatibility with Tibco Spotfire 12.0

• Expose xbar q function

  s)select xbar(10,x) from qt('([]1 12 23)')
  s)select xbar('0D00:10',x) from qt('([]0D+10:21 11:32 13:43)')